B10: Mobility and Activity Data from Smartphones
What Really Makes You Move? Identifying Relationships between Physical Activity and Health through Applying Machine Learning Techniques on High Frequency Accelerometer and Survey Data.
CentERdata - Tilburg University
Relevance & Research Question
Physical activity is an important indicator of health, but an accurate and objective measurement of physical activity is needed to gain insight and understanding of what drives differences in physical activity and how this influences health. Existing studies are generally based on self-report surveys and while the results of these studies are valuable, there are limitations to their use, e.g., varying perception of physical activity, social desirable answers, and incomplete recall of activity. Using wearable accelerometers as a measurement device provide a more complete and objective picture of physical activity and opens up new ways to study the relationship between physical activity and health. In addition, the influence of socioeconomics, -demographics and personality traits can be taken into account when studying these relationships.
Methods & Data
To study these relationships in detail, an experiment using accelerometers was conducted in the Dutch LISS panel. 1.000 panel members were invited to participate in the experiment. They were asked to wear an accelerometer for 8 days, measuring their physical activity level day and night. During this period, respondents regularly filled out surveys indicating what specific activities they conducted during the day. Furthermore, they provided details about their sedentary behavior, perceived health, the social context, and their associated mood.
First analyses of the objective accelerometer data and the subjective self-reported data showed how people tend to over-report physical activity compared to the objective measurements from the accelerometers. Moreover, these variations differ across socioeconomic and demographic groups (Kapteyn et al., 2018).
In a follow-up study we take the analyses of these high frequency data a step further by applying data science methods. Using machine learning techniques, such as deep (convolutional) neural networks for pattern recognition, clustering analyses, and classification, we are able to identify specific patterns, i.e. walking, running, cycling, sitting, sleeping. When we combine this with longitudinal survey data from the LISS panel we obtain relationships between physical activity and health on a detailed level. In essence, we gain insight in the relationship between the specific identified activities and personality traits, health, and socioeconomic and demographic status.
Squats in surveys: the use of accelerometers for fitness tasks in surveys
1Utrecht University; 2University of Mannheim; 3RECSM-Universitat Pompeu Fabra
Relevance & Research Question:
Smartphones are becoming increasingly important and widely-used in survey completion. Smartphones also offer many new possibilities for survey research, such as extending data collection by using sensor data (e.g., acceleration). Sensor data, for instance, can be used as a more objective supplement to health and physical fitness measures in mobile web surveys. In this study, we therefore investigate respondents’ willingness to participate in fitness tasks during mobile web survey completion. In addition, we investigate the appropriateness of acceleration data to draw conclusions about respondents’ health and fitness level.
Methods & Data:
59.3 % of the respondents expressed hypothetical willing to participate in a fitness task, 56.7% actually participated in the squat task. The acceleration data are currently being prepared for analyses, so results about its usefulness and comparability to self-reports of respondents’ health and fitness level will be available soon and will be presented at the GOR conference.
This study contributes to the development of more objective measures of respondents’ health and fitness in mobile web surveys and could be extended by further physical activity tasks in future research.
Marienthal 2.0: Research into the subtle effects of unemployment using smartphones
Institute for Employment Research (IAB), Germany
Unemployment has serious consequences for the lives of those affected. The classical Marienthal study from the 1930s already used innovative data collection methods to analyze the subtle effects of unemployment. The participating observation took on a special significance in the study design. Famous is the finding of a reduced walking speed with prolonged unemployment, caused by the loss of the daily structure.
Nowadays, smartphones provide researchers with new data sources to analyze the subtle effects of unemployment. Innovative tools, such as the sensors built into the smartphone, and at a previously unknown frequency provide unique data to update the Marienthal-findings.
In 2018, the Institute for Employment Research (IAB) conducted the IAB-SMART study. Respondents of the panel study “Labour market and social security” (PASS) were invited to install the IAB-SMART app on their Android smartphone. In addition to surveys, sensor data from the devices were collected via the app. Depending on the consent of the respondents, this included location information and data acceleration sensors or step counters. Similar to the Marienthal study, these passive measurements take place without conscious perception of the interviewees.
Smartphone data makes determining the mobility of the respondents in everyday life possible, such as the choice of means of transport, number of steps and also the respective speeds more precisely and without effort for (and with reduced influenceability by) the respondents. The link to PASS and the administrative data of the Federal Employment Agency allows this information to be linked to the labor market behavior of the persons.